|
| 1 | +package loadtest |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "fmt" |
| 6 | + "log/slog" |
| 7 | + "sync" |
| 8 | + "time" |
| 9 | + |
| 10 | + "github.com/google/uuid" |
| 11 | + "github.com/samber/lo" |
| 12 | + |
| 13 | + "github.com/castai/cluster-controller/internal/castai" |
| 14 | +) |
| 15 | + |
| 16 | +// CastAITestServer acts as simple cluster hub mock replacement. |
| 17 | +// It exposes a way to "push" actions to the cluster controller via GetActionsPushChannel |
| 18 | +// and can be used as an implementation of the server interface that cluster controller expects to call. |
| 19 | +type CastAITestServer struct { |
| 20 | + log *slog.Logger |
| 21 | + actionsPushChannel chan castai.ClusterAction |
| 22 | + cfg TestServerConfig |
| 23 | + |
| 24 | + logMx sync.Mutex |
| 25 | + actionsLog map[string]chan string |
| 26 | +} |
| 27 | + |
| 28 | +func NewTestServer(logger *slog.Logger, cfg TestServerConfig) *CastAITestServer { |
| 29 | + return &CastAITestServer{ |
| 30 | + log: logger, |
| 31 | + actionsPushChannel: make(chan castai.ClusterAction, 10000), |
| 32 | + cfg: cfg, |
| 33 | + actionsLog: make(map[string]chan string), |
| 34 | + } |
| 35 | +} |
| 36 | + |
| 37 | +// ExecuteActions pushes the list of actions to the queue for cluster controller to process. |
| 38 | +// This method returns when all actions are acked or context is cancelled. |
| 39 | +func (c *CastAITestServer) ExecuteActions(ctx context.Context, actions []castai.ClusterAction) { |
| 40 | + // owner channel has 1:n relationship with the actions. It handles the ack |
| 41 | + ownerChannel := make(chan string, len(actions)) |
| 42 | + |
| 43 | + for _, action := range actions { |
| 44 | + if action.ID == "" { |
| 45 | + action.ID = uuid.NewString() |
| 46 | + } |
| 47 | + c.addActionToStore(action.ID, ownerChannel) |
| 48 | + c.actionsPushChannel <- action |
| 49 | + } |
| 50 | + |
| 51 | + // Read from owner channel until len(actions) times, then close and return. |
| 52 | + finished := 0 |
| 53 | + for { |
| 54 | + select { |
| 55 | + case <-ctx.Done(): |
| 56 | + c.log.Info(fmt.Sprintf("Received signal to stop finished with cause (%q) and err (%v). Closing executor.", context.Cause(ctx), ctx.Err())) |
| 57 | + return |
| 58 | + case finishedAction := <-ownerChannel: |
| 59 | + c.removeActionFromStore(finishedAction) |
| 60 | + finished++ |
| 61 | + if finished == len(actions) { |
| 62 | + close(ownerChannel) |
| 63 | + return |
| 64 | + } |
| 65 | + } |
| 66 | + } |
| 67 | +} |
| 68 | + |
| 69 | +/* Start Cluster-hub mock implementation */ |
| 70 | + |
| 71 | +func (c *CastAITestServer) GetActions(ctx context.Context, _ string) ([]*castai.ClusterAction, error) { |
| 72 | + c.log.Info(fmt.Sprintf("GetActions called, have %d items in buffer", len(c.actionsPushChannel))) |
| 73 | + actionsToReturn := make([]*castai.ClusterAction, 0) |
| 74 | + |
| 75 | + // Wait for at least one action to arrive from whoever is pushing them. |
| 76 | + // If none arrive, we simulate the "empty poll" case of cluster-hub and return empty list. |
| 77 | + select { |
| 78 | + case x := <-c.actionsPushChannel: |
| 79 | + actionsToReturn = append(actionsToReturn, &x) |
| 80 | + case <-time.After(c.cfg.TimeoutWaitingForActions): |
| 81 | + c.log.Info(fmt.Sprintf("No actions to return in %v", c.cfg.TimeoutWaitingForActions)) |
| 82 | + return nil, nil |
| 83 | + case <-ctx.Done(): |
| 84 | + return nil, fmt.Errorf("context done with cause (%w), err (%w)", context.Cause(ctx), ctx.Err()) |
| 85 | + } |
| 86 | + |
| 87 | + // Attempt to drain up to max items from the channel. |
| 88 | + for len(actionsToReturn) < c.cfg.MaxActionsPerCall { |
| 89 | + select { |
| 90 | + case x := <-c.actionsPushChannel: |
| 91 | + actionsToReturn = append(actionsToReturn, &x) |
| 92 | + case <-time.After(50 * time.Millisecond): |
| 93 | + c.log.Info(fmt.Sprintf("Returning %d actions for processing", len(actionsToReturn))) |
| 94 | + // If we haven't received enough items, just flush. |
| 95 | + return actionsToReturn, nil |
| 96 | + case <-ctx.Done(): |
| 97 | + return nil, fmt.Errorf("context done with cause (%w), err (%w)", context.Cause(ctx), ctx.Err()) |
| 98 | + } |
| 99 | + } |
| 100 | + |
| 101 | + c.log.Info(fmt.Sprintf("Returning %d actions for processing", len(actionsToReturn))) |
| 102 | + return actionsToReturn, nil |
| 103 | +} |
| 104 | + |
| 105 | +func (c *CastAITestServer) AckAction(ctx context.Context, actionID string, req *castai.AckClusterActionRequest) error { |
| 106 | + errMsg := lo.FromPtr(req.Error) |
| 107 | + c.log.DebugContext(ctx, fmt.Sprintf("action %q acknowledged; has error: %v; error: %v", actionID, req.Error != nil, errMsg)) |
| 108 | + |
| 109 | + receiver := c.getActionReceiver(actionID) |
| 110 | + if receiver == nil { |
| 111 | + return fmt.Errorf("action %q does not have a receiver", actionID) |
| 112 | + } |
| 113 | + // Notify owner that this action was done. |
| 114 | + receiver <- actionID |
| 115 | + |
| 116 | + return nil |
| 117 | +} |
| 118 | + |
| 119 | +func (c *CastAITestServer) SendLog(ctx context.Context, e *castai.LogEntry) error { |
| 120 | + // No-op for now, maybe track metrics in the future? |
| 121 | + return nil |
| 122 | +} |
| 123 | + |
| 124 | +/* End Cluster-hub mock implementation */ |
| 125 | + |
| 126 | +func (c *CastAITestServer) addActionToStore(actionID string, receiver chan string) { |
| 127 | + c.logMx.Lock() |
| 128 | + defer c.logMx.Unlock() |
| 129 | + |
| 130 | + c.actionsLog[actionID] = receiver |
| 131 | +} |
| 132 | + |
| 133 | +func (c *CastAITestServer) removeActionFromStore(actionID string) { |
| 134 | + c.logMx.Lock() |
| 135 | + defer c.logMx.Unlock() |
| 136 | + |
| 137 | + delete(c.actionsLog, actionID) |
| 138 | +} |
| 139 | + |
| 140 | +func (c *CastAITestServer) getActionReceiver(actionID string) chan string { |
| 141 | + c.logMx.Lock() |
| 142 | + defer c.logMx.Unlock() |
| 143 | + |
| 144 | + receiver, ok := c.actionsLog[actionID] |
| 145 | + if !ok { |
| 146 | + c.log.Error(fmt.Sprintf("Receiver for action %s is no longer there, possibly shutting down", actionID)) |
| 147 | + return nil |
| 148 | + } |
| 149 | + return receiver |
| 150 | +} |
0 commit comments